Consideraciones sobre los recursos y ficha asociada

Table 5.8 shows potential independent variables which will be used in the severity models presented in Chapter 6. The weather condition variable was categorised as fine and 'other' which included fog, rain, dust, snow and cloud. After investigation it was decided to omit the weather variable because 99.67% of the crashes occurred during 'fine' weather conditions. In addition, after the development of an initial model the weather conditions variable was not significant in any models and therefore excluding it had little effect.

Table 5-8 Potential variables for severity models

Description of the variable Related statistics

Severity of the crash Slight injury=1, Serious Injury=2 and Fatal=3

Slight=3.82%, Serious=71.29%, and Fatal=24.90%.

Age of the driver (Years) A continuous variable Mean=30.21, Standard deviation=11.24

Road density (m/km2) A continuous variable Mean=14.45, Standard deviation=7.86

Nationality of the driver Non-Saudi=0; Saudi=1 Saudi=61.46% and Non- Saudi=38.54%

Time of day

00:00 to 03:59 Yes=15.56% and No=84.44%

04:00 to 07:59 Yes=17.08% and No=82.92%

08:00 to 11:59 Yes=20.16% and No=79.84%

12:00 to 15:59 Yes=15.17% and No=84.83%

16:00 to 19:59 Yes=17.16% and No=82.84%

20:00 to 23:59 Yes=10.22% and No=89.78%

Cause of the crash

Overtaking Overtaking=1, otherwise=0 Yes=33.02% and No=66.98% Distraction Distraction=1, otherwise=0 Yes=22.72% and No=77.28% Excessive speed Excessive speed=1, otherwise=0 Yes=16.39% and No=83.61%

103

Other (e.g. tiredness and sleeping) If the cause is Other=1, otherwise=0 Yes=27.87% and No=72.13% Type of collision

Angle angle=1, otherwise=0 Yes=15.81% and No=84.19%

Head-on Head-on=1, otherwise=0 Yes=10.63% and No=89.37% Rear-end If the type is Rear-end=1, otherwise=0 Yes=24.57% and No=75.43% Side-swipe Side-swipe=1, otherwise=0 Yes=16.50% and No=83.50% Other (unknown) Other=1, otherwise=0 Yes=27.82% and No=72.18% Location of the crash Straight=1, Other=0 Straight=88.01% and

Other=11.99%

Road surface Dry=1, Wet=0 Dry=99.67% and Wet=0.33%

Lighting condition of the road Lighting=1, without-lighting=0 Lighting=92% and Without lighting=8%

Weather Fine=1, Other=0 Fine=99.67% and other=0.33%

Single vehicle Single=1, Other=0 Single=43.27% and Other=56.73 Crash year

Year 1425 Yes=20.99% and No=79.01%

Year 1426 Yes=28.48% and No=71.52%

Year 1427 Yes=22.50% and No=77.50%

Year 1428 Yes=18.74% and No=81.74%

Year 1429 Yes=9.78% and No=90.22%

Number of casualties A continuous variable Mean=2.5282 Standard deviation=1.553

It is anticipated that road density (length of all roads per unit area) may also influence the severity of traffic crashes as vehicle average speeds are more likely to be low in areas with high road density owing to high traffic volumes and more pedestrian activity. Traffic crash data, however, do not include any information on road density in the vicinity of the crash location. Road density data are therefore calculated at spatial levels such as HAI, the equivalent of a ward, and geo-coded crash data are then integrated with the ward-level road density data using GIS. This allows ward-level road density data to be obtained for each of the crashes, which means that if a traffic crash occurs in a ward, the calculated road density of that ward is taken as the road density attributed to the crash.

In addition to these variables, interactions between age and nationality, age and excessive speed, and nationality and excessive speed will be tested in the models to be developed as shown in the next chapter.

A unique crash key was used to integrate these data files. They were collected over a period of five years, namely 1425, 1426, 1427, 1428, and 1429AH, roughly equivalent to 2004, 2005, 2006, 2007, and 2008, respectively. The severity of traffic crashes was

104

categorised as: fatal (2,498 crashes; 0.36%), serious (7,835 crashes; 1.12%), slight (1,127 crashes; 0.16%) and property damage only (686,785 crashes; 98.36%).

A dataset for use in the severity analysis was created containing only those cases where one driver has 100% responsibility, using the participant file; there were 573,206 cases. The sample with 100% of the blame was used because it has a single value for each variable (i.e. age and nationality for each participant), it shows the participant who is 100% responsible for the crash, and it shows all crash details. In the frequency analysis the full sample of the crash dataset is used.

Table 5.9 shows the dependent and independent variables which will be used in the frequency models presented in Chapter 7.

Table 5-9 Potential variables for frequency models

Variables Description

Dependent variables

Fatal crashes Number of fatal crashes in the HAI.

Serious injury crashes Number of serious injury crashes in the HAI. Demographic characteristics

Population Number of people living in the HAI.

Vehicle registered Number of vehicles in the crash, including all types (car, taxi, pickup/jeep, van, bus, truck).

% male Percentage of males in the HAI.

% non-Saudi Percentage of non-Saudi people in the HAI. Income per capita Total income divided by total population.

Income per adult >18 Total income divided by number of adult people >18 years old in the HAI % low income Percentage of people with income less than 2500 Saudi Riyals per month. % aged 0-18 Percentage of population from 0-18 years of age living in the HAI. % aged 65+ Percentage of people over 65 years in the population.

Employment Number of people in employment who live in the HAI. % illiterate Percentage of illiterate people in the HAI.

Road density Road density of each HAI estimated by dividing road lengths by the area of the HAI in km/km2.

Area Area or size of the HAI in km2. Land use

% residential Percentage of residential areas in the HAI.

% health utilities Percentage of health facilities areas in the HAI, such as clinics, hospitals (public and private), and GPs.

105

% educational utilities Percentage of educational utilities areas in the HAI, such as universities, colleges, institutes, and schools (primary, intermediate and high schools). % religious services percentage of religious areas in the HAI, such as mosques

% cultural utilities Percentage of cultural areas in the HAI, such as libraries, museums, exhibition halls, and any social gathering halls.

% agricultural Percentage of agricultural areas in the HAI, such as palm trees farms and all activities related to agriculture.

% industrial Percentage of industrial areas in the HAI.

% transport utilities Percentage of transport utilities areas in the HAI, such as car parks, transport stations in and outside the city, and taxis and limousines stations. % communications and

% public utilities

Percentage of communications and public utilities areas in the HAI, such as transmitting stations for land line telephones or wireless communications, radio and TV stations, and public utilities like electricity, gas, water, sewage disposal, and water treatment plants.

% recreation and parks Percentage of the recreation and parks areas in the HAI.

5.6 Summary

This chapter has discussed the data to be employed in the following chapters. This includes crash data involving details of the crash, participant data, and vehicle data. The data have been checked and validated for quality for fatal and serious injury crashes, but not for slight injury crashes, which were under-reported. As slight injury crashes were under-reported the analysis will be based on 1) fatal, serious injury, and slight injury crashes, and 2) binary (fatal and serious) which will be discussed in the next chapter.

Data were validated to ensure their accuracy and quality by performing statistical tests. It was concluded that the data used in this thesis are reliable and of good quality. The descriptive statistics of the variables such as number of crashes, age, and nationality in both crash severity and crash frequency models were also presented. Potential variables which will be used in severity analysis were also presented, as well as population land use and road network data which will be used in frequency analysis in Chapter 7.

106

6

CRASH SEVERITY MODELS

6.1 Introduction

The aim of this chapter, and one of the objectives of this thesis, is to identify the factors affecting the severity of traffic crashes that have occurred within Riyadh city; appropriate statistical models have been developed in an attempt to explore the relationship between crash severity and contributing factors. Crash severity refers to the level of severity of a road crash outcome (e.g. fatal, serious or slight).

In order to conduct this analysis, different econometric models were employed for modelling crash severity at the individual crash level. Crash severity is usually measured in categories (i.e. fatal, serious injury and slight injury). Therefore, an econometric model suitable for categorical data is needed.

Here two techniques are applied: • Ordered response models:

o Ordered logit model (OLM)

o Generalised ordered logit (GOLOGIT) model

o Partially constrained generalised ordered logit (PC-GOLOGIT) model • Unordered nominal response models:

o Multinomial logit (MNL) model o The mixed logit (ML) model

o The mixed binary logistic regression model.

In Table 5.8 in Chapter 5 are shown the potential independent variables which will be used in the severity models.

A number of hypotheses were formulated on the basis of the literature review in Chapter two and local knowledge:

- Younger drivers are associated with more severe crashes.

- Non-Saudi drivers are involved in more severe crashes as they are less familiar with local conditions.

107

- Single-vehicle crashes are more severe than multi-vehicle crashes. - Crash severity decreases over time.

- Time period 16:00 to 19:59 has more severe crashes as people go out for shopping and recreation.

When the sign of coefficient and t-statistics is positive, this means the crash is more likely to be severe, whereas when it is negative it is less likely.